Trajectory file with varying number of atoms

[Andrea]

Hello, We are three prospective AMP users and we were wondering if there was a way to load as trajectory files, snapshots composed of varying number of atoms, as it seems from ASE that this can be only done with a database module.

Thanks, G & A & R

[andrew_peterson]

Yes, that's a bit of a limitation of the ASE trajectory format. The database is a good way to proceed. If you want to avoid that format, you can just feed Amp a list of images that you assemble in your script. E.g., something like

images1 = ase.io.read('images1.traj', ':')
images2 = ase.io.read('images2.traj', ':')
images = images1 + images2

(then pass images to Amp.train). Does that work?

If you have suggestions for a better way, please let us know! We are trying to avoid creating yet another file format for atoms.

[Efrem Braun]

I don't think that ASE trajectory files have to have a constant number of atoms. Each of the "Atoms" objects in the trajectory is independent. My Amp script does the following (quite similar to what Andrew wrote above):

# Make training set trajectory
traj_train = Trajectory('training.traj', mode='w')
training_set_1 = read('input_1.xyz', index=(':'), format='xyz')
for atoms in training_set_1:
    traj_train.write(atoms)
training_set_2 = read('input_2.xyz', index=(':'), format='xyz')
for atoms in training_set_2:
    traj_train.write(atoms)
calc.train(images='training.traj')

I think the issue can be closed.

[andrew_peterson]

Yes, this was fixed upstream in ASE. The trajectory format now can take variable numbers of atoms.